CN104636750B - A kind of pavement crack recognizer and system based on double scale clustering algorithms - Google Patents

Abstract

The invention discloses a pavement crack recognition algorithm and system based on dual-scale clustering: the computer reads the three-dimensional image data matrix to obtain a binarized image; according to the sequence from top to bottom and from left to right, eight neighborhoods are used The search algorithm scans the data matrix corresponding to the binary image to obtain the marked crack area; in the ellipse corresponding to the crack area, use the dual-scale clustering algorithm to perform crack clustering to obtain the clustered crack area; use the clustered crack area The smallest circumscribing ellipse where the region is located is used as the pavement crack. The invention has low complexity, short running time and no manual participation. The idea of linear fitting and model building is used to represent the messy crack data locally into a regular and definite mathematical expression, thereby reducing the complexity of data processing; only need to input the collected pavement crack data to complete the pavement crack Therefore, the algorithm has high detection efficiency and fast speed, and has certain research value.

Description

A Pavement Crack Identification Algorithm and System Based on Dual-Scale Clustering Algorithm

technical field

This patent belongs to the field of road engineering, and in particular refers to a pavement crack identification algorithm based on a dual-scale clustering algorithm.

Background technique

The traditional pavement crack recognition technology is to carry out a series of image processing closely on the cracks and continuously optimize the effect to facilitate target extraction, that is, this type of algorithm is mainly dedicated to the extraction of crack features, and rarely considers the actual needs and processing. Whether the subsequent pavement cracks really correspond to the same actual pavement crack. First, the traditional crack feature extraction algorithm makes the preprocessed pavement crack binary image always reduce the actual pavement crack area to varying degrees compared with the original crack, and what's more, after multiple processing, the actual pavement crack area is reduced to a certain extent. Cracks appear to be fractured on the image. If a crack is not discovered and repaired in time, it is very likely to be treated as two or more cracks. Since the pavement crack identification process is a serial process, its There is a phenomenon of accumulation of errors, which makes the work of crack location in subsequent processing based on the results of front-end errors, which will inevitably produce wrong crack identification results, which greatly increases the error rate of crack identification, that is, there is a phenomenon of "if you don't make up, you will make mistakes". Second, in practice, the purpose of pavement crack detection is to accurately classify cracks, accurately locate cracks, and locate the actual area of cracks, so as to provide reliable data for road maintenance departments to facilitate road maintenance management.

Clustering algorithms have great application prospects in the field of image segmentation. It not only has great advantages in processing large amounts of data, but also has excellent scalability, which is convenient for discovering new research methods from different perspectives. Different clustering algorithms can be divided into "hard" clustering and "soft" clustering according to their different clustering criteria. As far as simple "hard" clustering is concerned, let the set C represent the image gray value data set, and cluster analysis on it is equivalent to dividing it into sub-regions c ₁ , c ₂ ..... ..c _k ,, k is the number of categories. Make the subregion satisfy the condition: non-emptiness: ;Completeness: c ₁ ∪c ₂ ∪c ₃ ∪…∪c _k ＝C, the essence of clustering algorithm is to redistribute the original data. The allocated data embodies some kind of internal consistency, which is usually measured by a specific criterion function. Using different criterion functions will result in different results. Optimizing the criterion function is one direction of optimizing the clustering algorithm. Commonly used clustering algorithms include hierarchical clustering algorithm, hybrid analytical mode query algorithm, nearest neighbor clustering algorithm, fuzzy clustering algorithm, artificial neural network clustering algorithm, genetic clustering algorithm, etc. Looking at these clustering algorithms, the core lies in the expression of "distance". The "distance" under different criteria reflects different clustering effects. Of course, for different data sets, different criteria can be selected accordingly to achieve a more ideal effect. Therefore, the definition of "distance" is particularly critical. At the same time, as far as pavement crack detection is concerned, clustering algorithms are rarely used in this field, and the only application is limited to the segmentation of pavement crack images, and there is no detection of crack area location.

Contents of the invention

Aiming at the problems existing in the above pavement crack identification technology, the present invention proposes a pavement crack identification algorithm based on dual-scale clustering, adopting the idea of "dividing first and then clustering", that is, performing local research on the cracks first, and then using the optimized center distance And the dual-scale clustering criterion of angle difference is used for dual-scale clustering of small fracture cracks. Finally, the minimum circumscribed ellipse model is used to characterize the cracks, so as to realize the location of cracks and the definition of regions, so as to identify pavement cracks. It not only avoids the implementation difficulties caused by the large amount of data and high complexity caused by calculating the manifold distance, but also achieves the advantages brought by the use of manifold distance clustering.

In order to achieve the above object, the present invention adopts following technical scheme:

A pavement crack identification algorithm based on dual-scale clustering, comprising the following steps:

Step 1: The computer reads the three-dimensional image data matrix, performs filtering processing on the three-dimensional image data matrix to obtain the filtered three-dimensional image data matrix, and then converts it into a grayscale image, and uses the Otsu algorithm to binarize the grayscale image Process to obtain a binarized image;

Step 2: In the order from top to bottom and from left to right, use the eight-neighborhood search algorithm to scan the data matrix corresponding to the binary image obtained in step 1 to obtain the marked crack area, and use The minimum circumscribing ellipse is used for characterization (that is, the crack area is enclosed by the minimum circumscribing ellipse);

Step 3: In the ellipse corresponding to the crack area, use the dual-scale clustering algorithm to cluster the cracks to obtain the clustered crack area;

Step 4: Use the smallest circumscribing ellipse where the clustered crack area is located as the pavement crack.

Further, it is characterized in that the step 2 specifically includes the following steps:

Step 21: Scan the filtered 3D image data matrix in order from top to bottom and from left to right; take each data point as the center in turn, and judge whether the eight neighborhoods of the center point are crack points, and if so, then The points within the eight neighborhoods belong to the basic unit area with the data point as the center point, and each basic unit area is numbered (marked) in sequence, and each basic unit area is used as a crack area;

Step 22: Perform linear fitting on all data points in the fracture area with number i to obtain a fitted straight line Among them, i=1,2,3...; n _i is the number of data points contained in the crack area of number i; find out the length of the line segment intercepted by the crack area on the corresponding fitting straight line, and record it as a _i ; then calculate The distance between all the data points in the fracture area and the corresponding fitting straight line, the maximum value of the distance is recorded as b _i , and the angle between a _i and the horizontal direction is calculated, which is recorded as θ _i ;

Step 23: Taking the parameters a _i , b _i and θ _i of each crack area as the major axis, minor axis and deflection angle of the corresponding ellipse, respectively, to obtain the minimum circumscribed ellipse corresponding to each crack area.

Further, the step 3 specifically includes the following steps:

Step 31: For all the crack regions obtained in step 2, select the center of the ellipse with the longest major axis among the smallest circumscribed ellipses corresponding to the crack region as the current clustering center;

Step 32: Select the smallest circumscribing ellipse where the fracture region with the smallest number among all fracture regions other than the fracture region corresponding to the current clustering center is selected as the target to be clustered;

Step 33: Calculate the distance from the center of the target to be clustered to the current cluster center and the horizontal angle difference between the two centers;

Step 34: Judging whether the distance obtained in step 33 and the horizontal angle difference satisfy the following criterion function, if not, go to step 37, and if so, execute step 35;

Criterion function: J＝(a ₀ <O ₀ O _i )&&(Δθ _oi <δ)

Among them, a ₀ is the major axis length of the smallest circumscribing ellipse represented by the current clustering center, O ₀ O _i is the distance from the center of the target to be clustered to the current clustering center, Δθ _oi is the distance between the target to be clustered and the current clustering center The horizontal angle difference of the class center, _δ1 is a normal number, taking 0-45°;

Step 35: classify the object to be clustered into the class where the current cluster center is located, and use a new minimum circumscribed ellipse to characterize all ellipses contained in the class (that is, use a minimum circumscribed ellipse for all ellipses), Calculate the center, major axis, minor axis and deflection angle of the new minimum circumscribed ellipse, and use the center of the new minimum circumscribed ellipse as the current clustering center;

Step 36: Determine whether there are unclassified crack regions, if so, continue to select the smallest circumscribing ellipse represented by the next smallest number as the target to be clustered, and return to step 33; if not, execute step 38;

Step 37: classify the current cluster center into a new class, return to 31, otherwise, execute step 38;

Step 38: End.

Further, in the step 1, the median filtering algorithm is used for filtering the three-dimensional image data matrix, and the Otsu algorithm is used for binarizing the grayscale image.

Another object of the present invention is to provide a pavement crack identification system based on dual-scale clustering, which system includes the following modules connected in sequence:

The image binarization module is used to complete the following functions:

The computer reads the three-dimensional image data matrix, performs filtering processing on the three-dimensional image data matrix to obtain the filtered three-dimensional image data matrix, and then converts it into a grayscale image, and uses the Otsu algorithm to binarize the grayscale image to obtain binarized image;

The crack area marking module is used to complete the following functions:

According to the order from top to bottom and from left to right, the eight-neighborhood search algorithm is used to scan the data matrix corresponding to the binary image obtained in step 1 to obtain the marked fracture area, and use the minimum circumscribed ellipse for each fracture area To characterize;

The fracture clustering module is used for the ellipse corresponding to the fracture area, and uses the dual-scale clustering algorithm to perform fracture clustering to obtain the clustered fracture area;

The pavement crack extraction module is used to use the smallest circumscribing ellipse where the clustered crack area is located as a pavement crack.

Further, the fracture area marking module is used to realize the following functions:

Step 21: Scan the filtered 3D image data matrix in order from top to bottom and from left to right; take each data point as the center in turn, and judge whether the eight neighborhoods of the center point are crack points, and if so, then The points within the eight neighborhoods belong to the basic unit area with the data point as the center point, and each basic unit area is numbered (marked) in sequence, and each basic unit area is used as a crack area;

Step 22: Perform linear fitting on all data points in the fracture area with number i to obtain a fitted straight line Among them, i=1,2,3...; n _i is the number of data points contained in the crack area of number i; find out the length of the line segment intercepted by the crack area on the corresponding fitting straight line, and record it as a _i ; then calculate The distance between all the data points in the fracture area and the corresponding fitting straight line, the maximum value of the distance is recorded as b _i , and the angle between a _i and the horizontal direction is calculated, which is recorded as θ _i ;

Step 23: Taking the parameters a _i , b _i and θ _i of each crack area as the major axis, minor axis and deflection angle of the corresponding ellipse, respectively, to obtain the minimum circumscribed ellipse corresponding to each crack area.

Further, the fracture clustering module is used to realize the following process functions:

Step 31: For all the crack regions obtained in step 2, select the center of the ellipse with the longest major axis among the smallest circumscribed ellipses corresponding to the crack region as the current clustering center;

Step 32: Select the smallest circumscribing ellipse where the fracture region with the smallest number among all fracture regions other than the fracture region corresponding to the current clustering center is selected as the target to be clustered;

Step 33: Calculate the distance from the center of the target to be clustered to the current cluster center and the horizontal angle difference between the two centers;

Step 34: Judging whether the distance obtained in step 33 and the horizontal angle difference satisfy the following criterion function, if not, go to step 37, and if so, execute step 35;

Criterion function: J＝(a ₀ <O ₀ O _i )&&(Δθ _oi <δ)

Among them, a ₀ is the major axis length of the smallest circumscribing ellipse represented by the current clustering center, O ₀ O _i is the distance from the center of the target to be clustered to the current clustering center, Δθ _oi is the distance between the target to be clustered and the current clustering center The horizontal angle difference of the class center, _δ1 is a normal number, taking 0-45°;

Step 35: classify the object to be clustered into the class where the current cluster center is located, and use a new minimum circumscribed ellipse to characterize all ellipses contained in the class (that is, use a minimum circumscribed ellipse for all ellipses), Calculate the center, major axis, minor axis and deflection angle of the new minimum circumscribed ellipse, and use the center of the new minimum circumscribed ellipse as the current clustering center;

Step 36: Determine whether there are unclassified crack regions, if so, continue to select the smallest circumscribing ellipse represented by the next smallest number as the target to be clustered, and return to step 33; if not, execute step 38;

Step 37: classify the current cluster center into a new class, return to 31, otherwise, execute step 38;

Step 38: End.

Further, in the image binarization module, the median filtering algorithm is used for filtering the three-dimensional image data matrix, and the Otsu algorithm is used for the binarization processing of the grayscale image.

The pavement crack identification technology proposed by the present invention has the following advantages

1. The detection of pavement cracks can be completed only by inputting the collected pavement crack data, so the algorithm is simple to calculate and has a short running time, which is suitable for use in real-time systems.

2. It does not require manual participation, and overcomes the shortcomings of manual pavement crack detection, such as high labor intensity, poor portability, low work efficiency and poor filtering effect.

3. On the basis of digital image processing technology, use the unique advantages of clustering algorithm to process big data, comprehensively analyze pavement crack images from a new perspective, and complete the dual-scale clustering algorithm identification of cracks.

4. Through the method of establishing a mathematical model of the crack area, the chaotic pavement crack area is expressed as a formula that can be specifically expressed by mathematical formulas, which is convenient for analysis using mathematical techniques. It provides powerful information support for the maintenance and management of the road surface, and improves the level of road maintenance and management.

Description of drawings

FIG. 1 is a flow chart of the dual-scale clustering-based pavement crack identification algorithm of the present invention.

Fig. 2 is the result of using the present invention to treat three kinds of cracks respectively. Among them, (a), (d), (g) are the original transverse cracks, original longitudinal cracks and original network cracks; (b), (e), (h) are (a), (d), ( The crack basic unit of g); (c), (f), and (i) are the results of (a), (d), and (g) processed by the algorithm of the present invention in turn.

Fig. 3 is a schematic diagram of the pavement crack identification system based on dual-scale clustering of the present invention.

The implementation effect of the pavement crack identification technology of the present invention is illustrated below in conjunction with examples:

Detailed ways

Referring to Fig. 1-Fig. 3, the pavement crack recognition algorithm based on dual-scale clustering algorithm of the present invention specifically includes the following steps:

Step 1: The computer reads the 3D image data matrix, and uses the median filter algorithm to filter the 3D image data matrix to obtain the filtered 3D image data matrix, then converts it into a grayscale image, and uses the grayscale image The Otsu algorithm performs binarization processing to obtain a binarized image;

In this step, the use of the median filter algorithm has low requirements on the smoothness of the data, so it is suitable for the processing of staggered data in the present invention, thereby improving the operation speed. At the same time, the Otsu algorithm is an adaptive threshold algorithm with a high degree of automation.

Step 2: In the order from top to bottom and from left to right, use the eight-neighborhood search algorithm to scan the data matrix corresponding to the binary image obtained in step 1 to obtain the marked crack area, and use The minimum circumscribing ellipse is used for characterization (that is, the crack area is enclosed by the minimum circumscribing ellipse);

Specifically include the following steps:

Step 21: Scan the filtered 3D image data matrix in order from top to bottom and from left to right; take each data point as the center in turn, and judge whether the eight neighborhoods of the center point are crack points, and if so, then The points within the eight neighborhoods belong to the basic unit area with the data point as the center point, and each basic unit area is numbered (marked) in sequence, and each basic unit area is used as a crack area;

Step 22: Perform linear fitting on all data points in the fracture area with number i to obtain a fitted straight line Among them, i=1,2,3...; n _i is the number of data points contained in the crack area of number i; find out the length of the line segment intercepted by the crack area on the corresponding fitting straight line, and record it as a _i ; then calculate The distance between all the data points in the fracture area and the corresponding fitting straight line, the maximum value of the distance is recorded as b _i , and the angle between a _i and the horizontal direction is calculated, which is recorded as θ _i ;

Step 23: Taking the parameters a _i , b _i and θ _i of each crack area as the major axis, minor axis and deflection angle of the corresponding ellipse, respectively, to obtain the minimum circumscribed ellipse corresponding to each crack area.

In this step, the eight-neighborhood search algorithm is used to search comprehensively, making the result more accurate; at the same time, using the minimum circumscribed ellipse for representation is more accurate for crack representation than the existing rectangle representation.

Step 3: In the ellipse corresponding to the crack area, use the dual-scale clustering algorithm to cluster the cracks to obtain the clustered crack area; specifically include the following steps:

Step 31: For all the crack regions obtained in step 2, select the center of the ellipse with the longest major axis among the smallest circumscribed ellipses corresponding to the crack region as the current clustering center;

Step 32: Select the smallest circumscribing ellipse where the fracture region with the smallest number among all fracture regions other than the fracture region corresponding to the current clustering center is selected as the target to be clustered;

Step 33: Calculate the distance from the center of the target to be clustered to the current cluster center and the horizontal angle difference between the two centers;

Step 34: Judging whether the distance obtained in step 33 and the horizontal angle difference satisfy the following criterion function, if not, go to step 37, and if so, execute step 35;

Criterion function: J＝(a ₀ <O ₀ O _i )&&(Δθ _oi <δ)

Among them, a ₀ is the major axis length of the smallest circumscribing ellipse represented by the current clustering center, O ₀ O _i is the distance from the center of the target to be clustered to the current clustering center, Δθ _oi is the distance between the target to be clustered and the current clustering center The horizontal angle difference of the class center, _δ1 is a normal number, taking 0-45°;

Step 35: classify the object to be clustered into the class where the current cluster center is located, and use a new minimum circumscribed ellipse to characterize all ellipses contained in the class (that is, use a minimum circumscribed ellipse for all ellipses), Calculate the center, major axis, minor axis and deflection angle of the new minimum circumscribed ellipse, and use the center of the new minimum circumscribed ellipse as the current clustering center;

Step 36: Determine whether there are unclassified crack regions, if so, continue to select the smallest circumscribing ellipse represented by the next smallest number as the target to be clustered, and return to step 33; if not, execute step 38;

Step 37: classify the current cluster center into a new class, return to 31, otherwise, execute step 38;

Step 38: End.

The selection of the criterion function in this step can not only judge the distance but also judge the angle, can judge the error comprehensively, and improve the accuracy of the result.

Step 4: Use the smallest circumscribing ellipse where the clustered crack area obtained in Step 3 is used as the pavement crack to realize the location detection of the pavement crack.

Referring to Fig. 2, in order to identify pavement cracks using the dual-scale clustering algorithm, the experiment selects δ = 30°. It can be seen that the use of dual-scale clustering algorithm can obtain accurate fracture range and fracture information extraction, so as to complete the detection of fractures.

Claims (4)

1. a kind of pavement crack recognition methods based on double scales cluster, which is characterized in that include the following steps：

Step 1：Computer reads 3 d image data matrix, is filtered after obtaining filtering to 3 d image data matrix 3 d image data matrix, be then converted into gray level image, and two-value is carried out using Otsu algorithms to the gray level image Change is handled, and obtains binary image；

Step 2：According to sequence from top to bottom, from left to right, the binaryzation obtained using eight neighborhood searching algorithm scanning step 1 The corresponding data matrix of image, the crack area after being marked, and minimum external oval progress is used to each crack area Characterization；The step 2 specifically comprises the following steps：

Step 21：According to the corresponding data matrix of the filtered binary image of sequential scan from top to bottom, from left to right；According to It is secondary centered on each data point, judge whether the eight neighborhood of central point is crack point, if it is, within the scope of eight neighborhood Point belong to the basic unit region put centered on the data point, and each basic unit region is sequentially numbered, Mei Geji This unit area is used as crack area；

Step 22：Linear fit is carried out to all data points in the crack area of number i, obtains fitting a straight lineWherein, i=1,2,3 ...；n_iThe number that crack area for number i includes Strong point number；The line segment length that the crack area intercepts in corresponding fitting a straight line is found out, a is denoted as_i；Then the crack is calculated The maximum value of distance is denoted as b by all data points to the distance of corresponding fitting a straight line in region_i, calculate a_iWith horizontal direction Angle is denoted as θ_i；

Step 23：By the parameter a of each crack area_i、b_iAnd θ_iRespectively as corresponding elliptical long axis, short axle and deflection Angle obtains the corresponding external ellipse of minimum of each crack area；

Step 3：In the corresponding ellipse of crack area, crack cluster, splitting after being clustered are carried out using double scale clustering algorithms Stitch region；The step 3 specifically comprises the following steps：

Step 31：For all slits region that step 2 obtains, long axis is chosen in the external ellipse of corresponding minimum of crack area most Long elliptical center is as current cluster centre；

Step 32：It chooses and numbers minimum crack area in all slits region except the corresponding crack area of current cluster centre The external ellipse of minimum where domain, as target to be clustered；

Step 33：The center for calculating target to be clustered is poor to the distance of current cluster centre and the horizontal sextant angle at two centers；

Step 34：Whether the distance and horizontal sextant angle difference that judgment step 33 obtains meet following criterion function, if not satisfied, then Step 37 is gone to, if it is satisfied, then executing step 35；

Criterion function：J=(a₀<O₀O_i)&&(Δθ_oi<δ₁)

Wherein, a₀For the representative external elliptical long axial length of minimum of current cluster centre, O₀O_iFor the center of target to be clustered To the distance of current cluster centre, Δ θ_oiPoor, the δ for target to be clustered and the horizontal sextant angle of current cluster centre₁For normal number, take 0~45 °；

Step 35：The class target to be clustered being classified as where current cluster centre, all oval uses that will include in such One new external ellipse of minimum is characterized, and the new external elliptical center of minimum, long axis, short axle and the deflection is calculated Angle, and using the new external elliptical center of minimum as current cluster centre；

Step 36：Judge whether also unclassified crack area, if so, then continuing to choose next lowest number representative most Small external ellipse is used as target to be clustered, and return to step 33；If no, executing step 38；

Step 37：Current cluster centre is classified as a new class, returns to 31, otherwise, executes step 38；

Step 38：Terminate；

Step 4：It is external oval as pavement crack using the minimum where the crack area after cluster.

2. the pavement crack recognition methods as described in claim 1 based on double scales cluster, which is characterized in that the step 1 In, 3 d image data matrix is filtered using median filtering algorithm.

3. a kind of pavement crack identifying system based on double scales cluster, which is characterized in that including being sequentially connected the mould connect as follows Block：

Image binaryzation module, to complete following function：

Computer reads 3 d image data matrix, is filtered to obtain filtered three-dimensional to 3 d image data matrix Image data matrix is then converted into gray level image, and carries out binary conversion treatment using Otsu algorithms to the gray level image, Obtain binary image；

Crack area mark module, to complete following function：

According to sequence from top to bottom, from left to right, two obtained using eight neighborhood searching algorithm scan image binarization block The corresponding data matrix of value image, the crack area after being marked, and minimum external ellipse is used to each crack area It is characterized；

The function of the crack area mark module specific implementation following below scheme：

Step 21：According to the corresponding data matrix of the filtered binary image of sequential scan from top to bottom, from left to right；According to It is secondary centered on each data point, judge whether the eight neighborhood of central point is crack point, if it is, within the scope of eight neighborhood Point belong to the basic unit region put centered on the data point, and each basic unit region is sequentially numbered, Mei Geji This unit area is used as crack area；

Step 22：Linear fit is carried out to all data points in the crack area of number i, obtains fitting a straight lineWherein, i=1,2,3 ...；n_iThe number that crack area for number i includes Strong point number；The line segment length that the crack area intercepts in corresponding fitting a straight line is found out, a is denoted as_i；Then the crack is calculated The maximum value of distance is denoted as b by all data points to the distance of corresponding fitting a straight line in region_i, calculate a_iWith horizontal direction Angle is denoted as θ_i；

Step 23：By the parameter a of each crack area_i、b_iAnd θ_iRespectively as corresponding elliptical long axis, short axle and deflection Angle obtains the corresponding external ellipse of minimum of each crack area；

Crack cluster module, in the corresponding ellipse of crack area, to carry out crack cluster using double scale clustering algorithms, obtain Crack area after cluster；The crack cluster module is realizing the function of following flow：

Step 31：For all slits region that step 2 obtains, long axis is chosen in the external ellipse of corresponding minimum of crack area most Long elliptical center is as current cluster centre；

Step 32：It chooses and numbers minimum crack area in all slits region except the corresponding crack area of current cluster centre The external ellipse of minimum where domain, as target to be clustered；

Step 33：The center for calculating target to be clustered is poor to the distance of current cluster centre and the horizontal sextant angle at two centers；

Step 34：Whether the distance and horizontal sextant angle difference that judgment step 33 obtains meet following criterion function, if not satisfied, then Step 37 is gone to, if it is satisfied, then executing step 35；

Criterion function：J=(a₀<O₀O_i)&&(Δθ_oi<δ₁)

Wherein, a₀For the representative external elliptical long axial length of minimum of current cluster centre, O₀O_iFor the center of target to be clustered To the distance of current cluster centre, Δ θ_oiPoor, the δ for target to be clustered and the horizontal sextant angle of current cluster centre₁For normal number, take 0~45 °；

Step 35：The class target to be clustered being classified as where current cluster centre, all oval uses that will include in such One new external ellipse of minimum is characterized, and the new external elliptical center of minimum, long axis, short axle and the deflection is calculated Angle, and using the new external elliptical center of minimum as current cluster centre；

Step 36：Judge whether also unclassified crack area, if so, then continuing to choose next lowest number representative most Small external ellipse is used as target to be clustered, and return to step 33；If no, executing step 38；

Step 37：Current cluster centre is classified as a new class, returns to 31, otherwise, executes step 38；

Step 38：Terminate；

Pavement crack extraction module oval is used as pavement crack to the minimum where the crack area after clustering is external.

4. the pavement crack identifying system as claimed in claim 3 based on double scales cluster, which is characterized in that described image two In value module, 3 d image data matrix is filtered using median filtering algorithm.